NeuralNote is an audio plugin that transcribes audio to MIDI using deep learning models.
The tool solves the problem of manually transcribing audio recordings into MIDI notation, which is time-consuming and error-prone. It works by applying a transformer-based transcription model to audio input, either from live recording or from dropped audio files. The current version uses MuScriptor, a model with 103M to 1.4B parameters depending on configuration, which can identify multiple instruments in a mix and output separate MIDI tracks for each. Transcription runs locally on the user's machine using GPU or CPU, and audio never leaves the system except when downloading models or checking for updates.
The tool suits music producers, composers, and musicians who need to convert existing recordings into editable MIDI form. It works as a VST3 or AU plugin in Digital Audio Workstations, or as a standalone application. The plugin accepts audio via live recording or by dropping files in common formats including WAV, AIFF, FLAC, MP3, and OGG Vorbis. A built-in multi-instrument synthesizer allows listening to transcribed results with per-instrument gain, mute, and solo controls. Users can export transcribed MIDI by dragging it to a MIDI track or saving to disk. The tool differs from its predecessor by replacing a smaller model with a significantly larger transformer architecture and adding multi-instrument transcription capability, whereas earlier versions transcribed one instrument at a time.
The project maintains active engagement with users through GitHub issues for hardware compatibility reporting, acknowledging that testing coverage remains limited across different machines and GPUs. Development has produced a major version update introducing a substantially different transcription model and architectural changes to the plugin's capabilities. The tool is distributed as signed and notarized installers for macOS with Apple Silicon and as unsigned installers for Windows x64, with prebuilt binaries for Linux listed as forthcoming.
